Improved CEEMDAN Based Speech Signal Analysis Algorithm for Mental Disorders Diagnostic System

Abstract

An automated algorithm for pitch frequency measurement for diagnostic systems of borderline mental disorders is developed. It is based on decomposition of a speech signal into frequency components using an adaptive method for analyzing of non-stationary signals, improved complete ensemble empirical mode decomposition with adaptive noise (improved CEEMDAN), and isolating the component containing pitch. A block diagram for the developed algorithm and a detailed mathematical description are presented. A research of the algorithm using the formed verified signal base of healthy patients, and male and female patients with psychogenic disorders, aged from 18 to 60, is conducted. The research results are evaluated in comparison with the known algorithms for pitch frequency measurement. In accordance with the results of the study, the developed algorithm for pitch frequency measurement provides an accuracy increase in determination of borderline mental disorders: for the error of the first kind, on the average, it is more accurate by 10.7%, and for the second type error by 4.7%.